JP7065185B2 - Electrocardiographic signal detection - Google Patents

Description

The present invention relates to the field of electronic technology, and particularly to electrocardiographic signal detection methods, devices and electronic devices.

In recent years, with the rise of deep learning methods, more and more researchers have begun to adopt modes to train neural networks to classify and recognize electrocardiographic (ECG) signals. However, while most of them recognize and classify electrocardiographic signals for a single heartbeat, none have yet performed pathological recognition for consecutive ECG polycardiacs.

In view of this, the present invention provides a new solution that allows pathological diagnosis to be performed on continuous electrocardiographic signals.

In order to achieve the above object, the solution according to the present invention is as follows.

The first aspect of the present invention provides a method for detecting an electrocardiographic signal. The electrocardiographic signal detection method is
A step of dividing an electrocardiographic signal of a predetermined time length to obtain a first predetermined number of single heartbeats, and
A step of identifying the feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats and acquiring the feature data of the first predetermined number of heartbeats.
A step of identifying the pathological type of the electrocardiographic signal of the predetermined time length based on the electrocardiographic signal of the predetermined time length and the characteristic data of the first predetermined number is included.

A second aspect of the present invention provides a method for detecting an electrocardiographic signal. The electrocardiographic signal detection method is
The step of identifying the pathological type of the electrocardiographic signal of a predetermined time length by the second convolutional neural network, and
When the pathological type indicates that an abnormality has occurred in the electrocardiographic signal, a step of dividing the electrocardiographic signal having a predetermined time length to obtain a first predetermined number of single heartbeats and
By inputting the data of the first predetermined number of single heartbeats into the first convolutional neural network, the step of identifying the heartbeat position where the abnormality occurred in the first predetermined number of single heartbeats by the first convolutional neural network. And, including.

A third aspect of the present invention provides an electrocardiographic signal detection device. The electrocardiographic signal detection device is
A first division module for dividing an electrocardiographic signal having a predetermined time length to obtain a first predetermined number of single heartbeats,
A first specific module for specifying the feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats acquired by the first division module and acquiring the first predetermined number of feature data, and
A second for identifying the pathological type of the electrocardiographic signal having a predetermined time length based on the electrocardiographic signal having a predetermined time length and the feature data of the first predetermined number specified by the first specific module. It is equipped with a specific module.

A fourth aspect of the present invention provides an electrocardiographic signal detection device. The electrocardiographic signal detection device is
A fourth specific module for identifying the pathological type of an electrocardiographic signal of a predetermined time length by a second convolutional neural network, and
When the pathological type specified by the fourth specific module indicates that an abnormality has occurred in the electrocardiographic signal, the electrocardiographic signal having a predetermined time length is divided to obtain a first predetermined number of single heartbeats. 2nd split module for
By inputting the data of the first predetermined number of single heartbeats acquired by the second division module into the first convolutional neural network, the abnormality in the first predetermined number of single heartbeats by the first convolutional neural network It is provided with a fifth specific module for specifying the heartbeat position where the above occurs.

A fifth aspect of the present invention provides a device-readable storage medium. The device-readable storage medium stores an executable command of the device for executing the electrocardiographic signal detection method submitted in the first or second aspect.

A sixth aspect of the present invention provides an electronic device. The electronic device comprises a processor and a storage medium for storing an executable command of the processor, the processor performing the electrocardiographic signal detection method submitted in the first or second aspect. ..

As can be seen from the above-mentioned solutions, since any single heartbeat is not isolated in the ECG signal of the continuous time series but is related to the single heartbeats adjacent to each other before and after, the first predetermined number of single heartbeats in the present invention. The characteristic data corresponding to each single heartbeat of the above can very well represent the pathological characteristics of the representative electrocardiographic signal. Therefore, the pathological type of the electrocardiographic signal having a predetermined time length can be detected very well by the electrocardiographic signal and the first predetermined number of feature data.

It is a schematic flowchart of the electrocardiographic signal detection method of an exemplary embodiment of this invention.

It is a schematic diagram of the continuous electrocardiographic signal in the embodiment shown in FIG. 1A.

FIG. 3 is a schematic diagram of a single heartbeat in the embodiment shown in FIG. 1A.

It is a schematic flowchart of the electrocardiographic signal detection method of another exemplary embodiment of this invention.

FIG. 2 is a schematic diagram of an architecture for detecting an electrocardiographic signal applied to the embodiment shown in FIG. 2A.

It is a structural schematic diagram of the 1st convolutional neural network in the Example shown in FIG. 2A.

It is a structural schematic diagram of the 2nd convolutional neural network in the Example shown in FIG. 2A.

It is a schematic flowchart of the electrocardiographic signal detection method of still another exemplary embodiment of this invention.

It is one of the schematic diagrams of the architecture for detecting an electrocardiographic signal applied to the embodiment shown in FIG. 3A.

FIG. 2 is a schematic diagram of an architecture for detecting an electrocardiographic signal applied to the embodiment shown in FIG. 3A.

It is a schematic flowchart of the electrocardiographic signal detection method of another example embodiment of this invention.

FIG. 4 is a schematic diagram of an architecture for detecting an electrocardiographic signal applied to the embodiment shown in FIG. 4A.

It is structural schematic diagram of the electrocardiographic signal detection apparatus of an exemplary embodiment of this invention.

It is a structural schematic diagram of the electrocardiographic signal detection apparatus of another exemplary embodiment of this invention.

It is a structural schematic diagram of the electrocardiographic signal detection apparatus of still another exemplary embodiment of this invention.

It is a structural schematic diagram of the electronic device of an example example of this invention.

Here, exemplary embodiments will be described in detail. An example is shown in the drawings. The following description, when relating to a drawing, indicates elements of the same or similar elements in different drawings, unless otherwise indicated. The embodiments described in the following exemplary examples are not representative of all embodiments consistent with the present invention. Conversely, they are examples of devices and methods that are consistent with some aspects of the invention, as described in detail in the appended claims.

The terminology used in the present invention is solely for the purpose of describing a particular embodiment and is not intended to limit the invention. The singular forms "type", "above" and "corresponding" used in the present invention and the appended claims also include a plurality of forms unless the other meanings are clearly understood from the context. Intended. It should be understood that the term "and / or" used in the text refers to any or all possible combinations including one or more related enumeration items.

It should be understood that various information is described by using the terms first, second, third, etc. in the present invention, but these information are not limited to these terms. These terms are used simply to distinguish between the same type of information. For example, as long as it does not deviate from the scope of the present invention, the first information may be referred to as the second information, and similarly, the second information may be referred to as the first information. This depends on the context. For example, the word "case" used herein may be interpreted as "... when", "... when" or "according to a particular situation".

To further illustrate the invention, examples are provided below.

1A is a schematic flowchart of an electrocardiographic signal detection method according to an exemplary embodiment of the present invention, FIG. 1B is a schematic diagram of a continuous electrocardiographic signal according to the embodiment shown in FIG. 1A, and FIG. 1C. Is a schematic diagram of the ECG single heartbeat in the example shown in FIG. 1A. The present invention may be applied to electronic devices such as wearable devices and handheld devices, and monitors the user's heart health status. As shown in FIG. 1A, it includes the following steps.

In step 101, the electrocardiographic signal having a predetermined time length is divided to obtain a first predetermined number of single heartbeats.

単一心拍の持続時間に対して正規化を行う過程に、固定のサンプリングレートでサンプリングを行ってもよい。Ｐは、各サンプリング点での単一心拍の信号強度を示す。例えば、Ｐ_１は、第１個のサンプリング点での単一心拍の信号強度を示し、Ｐ_Ｌは、第Ｌ個のサンプリング点での単一心拍の信号強度を示し、ｐ_１、ｐ_２、・・・ｐ_Ｌは、Ｌ個のサンプリング点での単一心拍の信号強度を示し、ｔは、当該単一心拍の持続時間である。第１所定数がＮであることを例示として説明すると、連続するＮ個のＥＣＧ単一心拍は、下記のように表されてもよい。

Ｐ_ijは、第ｊ個のサンプリング点での第ｉ個の単一心拍の信号強度を示し、ｉ＝１、２、・・・Ｎ、ｊ＝１、２、・・・Ｌ、ｔ_１、ｔ_２、・・・ｔ_Ｎは、Ｎ個の単一心拍のそれぞれの持続時間を示す。 In one embodiment, the ECG signal having a predetermined time length may be divided according to the recognition method for the ECG signal to acquire a first predetermined number of single heartbeats. As shown in FIGS. 1B and 1C, by dividing a continuous ECG signal having a predetermined time length, the start time point, end time point, and duration of each single heartbeat can be grasped. Moreover, the duration t of a single heartbeat may be normalized to a predetermined length. For example, when the predetermined length is L, the single heartbeat may be expressed as follows.

Sampling may be performed at a fixed sampling rate during the normalization process for the duration of a single heartbeat. P indicates the signal strength of a single heartbeat at each sampling point. For example, P ₁ indicates the signal strength of a single heartbeat at the first sampling point, PL indicates the signal strength of a single heartbeat at the _Lth sampling point, p ₁ , p ₂ , p. ... p _L indicates the signal strength of a single heartbeat at L sampling points, and t is the duration of the single heartbeat. Explaining that the first predetermined number is N as an example, N consecutive ECG single heartbeats may be expressed as follows.

P _ij indicates the signal strength of the i-th single heartbeat at the j-th sampling point, i = 1, 2, ... N, j = 1, 2, ... L, t ₁ , t ₂ , ... t _N indicate the duration of each of N single heartbeats.

In step 102, the feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats is specified, and the feature data of the first predetermined number is acquired.

In one embodiment, feature data corresponding to each single heartbeat of a first predetermined number of single heartbeats may be specified based on a deep learning network. The feature data may be a single feature or a combination of a plurality of features. A single heartbeat signal may be input to a deep learning network, a convolution layer of the deep learning network may be set, a convolution process may be performed on the single heartbeat signal, and then feature data may be acquired.

Corresponding to the above step 101, the feature data is, for example, T ₁ , T ₂ , ... _TN .

That is, in step 102, _Pi → _Ti can be realized. Here, i = 1, 2, ... N.

In step 103, the pathological type of the electrocardiographic signal having a predetermined time length is specified based on the electrocardiographic signal having a predetermined time length and the feature data of the first predetermined number.

In one embodiment, an electrocardiographic signal having a predetermined time length is input to an input layer of one convolutional neural network, a first predetermined number of feature data is input to the convolutional layer of the convolutional neural network, and the processing of the convolutional neural network is performed. , Identify the pathological type of the electrocardiographic signal in the output layer of the convolutional neural network. In one embodiment, the convolutional neural network may be trained with a huge electrocardiographic signal having various different pathological features. By training the convolutional neural network, the convolutional neural network can accurately recognize the pathological type of the electrocardiographic signal.

In one embodiment, the pathological type may include atrial extrasystoles, ventricular extrasystoles, atrial fibrillation, atrial flutter, ventricular tachycardia, and the like. It should be explained that the pathology type is merely an exemplary description and does not constitute a limitation to the present invention.

Since no single heartbeat in the continuous ECG signal is isolated and is associated with adjacent single heartbeats before and after, the feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats in the present invention is available. , The pathological characteristics of the representative electrocardiographic signal can be expressed very well. Therefore, the pathological type of the electrocardiographic signal having a predetermined time length can be satisfactorily detected based on the electrocardiographic signal and the first predetermined number of feature data.

FIG. 2A is a schematic flowchart of an electrocardiographic signal detection method according to another exemplary embodiment of the present invention, and FIG. 2B is an architecture for detecting an electrocardiographic signal applied to the embodiment shown in FIG. 2A. It is a schematic diagram, FIG. 2C is a structural schematic diagram of the first convolutional neural network in the embodiment shown in FIG. 2A, and FIG. 2D is a structural schematic diagram of the second convolutional neural network in the embodiment shown in FIG. 2A. .. As shown in FIG. 2A, it includes the following steps.

In step 201, the electrocardiographic signal having a predetermined time length is divided to obtain a first predetermined number of single heartbeats.

Since the description of step 201 may refer to the description of the embodiment shown in FIG. 1A, it will not be described in detail here.

In step 202, the feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats is extracted by sequentially inputting the data of the first predetermined number of single heartbeats into the first convolutional neural network.

In step 203, the time-series data corresponding to each single heartbeat in the electrocardiographic signal having a predetermined time length is specified, and the first predetermined number of time-series data is acquired.

In step 204, the pathological type of the electrocardiographic signal is specified by inputting the first predetermined number of time series data and the first predetermined number of feature data to the input layer of the second convolutional neural network.

In step 205, a determination is made for each single heartbeat of the first predetermined number of single heartbeats by the first convolutional neural network, and the determination result is acquired.

In step 206, the heartbeat position where the abnormality occurs in the first predetermined number of single heartbeats is specified based on the determination result.

It should be explained that the steps 205 and 206 are not necessarily executed after step 204, and steps 205 and 206 may be executed after step 202. In this way, the first convolutional neural network can recognize the heartbeat position where the abnormality occurs in the first predetermined number of single heartbeats.

In the following, an exemplary description will be given to the present embodiment in combination with FIGS. 2B and 2D.

In step 202, as shown in FIG. 2B, 25 consecutive single heartbeats and the start time point, end time point, and duration of the 25 single heartbeats are acquired by the step 201, and the single heartbeat is obtained. Perform length normalization for each of the data in. For example, when the normalized length is 196, 25 single heartbeat data having a length of 196 are sequentially input to the first convolutional neural network, and each single is in a predetermined convolutional layer of the first convolutional neural network. Feature data corresponding to one heartbeat may be output. For example, if each single heartbeat corresponds to 5 * 2 feature data, 25 single heartbeats correspond to 5 * 2 * 25 feature data. As shown in FIG. 2B, after acquiring the feature data of each single heartbeat by the first convolutional neural network, the feature data corresponding to each single heartbeat may be cached.

Ｎ個の単一心拍は、以下のように表されてもよい。

Ｎは、第１所定数、即ち、単一心拍の数である。 In step 203 and step 204, in one embodiment, for each single heartbeat in an electrocardiographic signal having a predetermined time length, a time point corresponding to the R wave of the single heartbeat is specified, and the R of the single heartbeat is specified. The time point corresponding to each R wave of the second predetermined number of single heartbeats adjacent to the wave before and after is specified, the time point corresponding to the R wave of the single heartbeat and the second time point adjacent to the single heartbeat before and after. Time-series data corresponding to the single heartbeat may be specified based on the time point corresponding to each of the R waves of a predetermined number of single heartbeats. For example, if the second predetermined number is 2, the distance between the current single heartbeat R wave and the two front and rear single heartbeats (that is, a total of four single heartbeats) is x ₁ , respectively. It is x2, _{x3 ,} and _x4 . When the rhythm information X of each single heartbeat is shown as one five-dimensional vector, the following is satisfied.

N single heartbeats may be expressed as:

N is the first predetermined number, that is, the number of single heartbeats.

For example, when a single heartbeat corresponds to m1 * n1 feature data and m2 * n2 time series data, the amount of data input to the second convolutional neural network is expressed as m1 * n1 + m2 * n2. You may. As shown in FIG. 2B, the single heartbeat input to the second convolutional neural network corresponds to 5 * 2 * 25 feature data, and the single heartbeat input to the second convolutional neural network is 5 * 1 *. When corresponding to 25 time series data, the amount of data input to the second convolutional neural network may be expressed as 5 * 3 * 25.

In step 204 above, after the second convolutional neural network is trained by a huge electrocardiographic signal with various pathological features, the second convolutional neural network is based on the data of the input layer and the pathological type of the electrocardiographic signal. Can be accurately recognized.

In steps 205 and 206 above, the first convolutional neural network may be trained with an enormous normal and abnormal single heartbeat electrocardiogram. After training, the first convolutional neural network can accurately recognize whether or not an abnormality has occurred in a single heartbeat. For example, after 25 single heartbeats are sequentially input to the first convolutional neural network, the first convolutional neural network may make a determination for normality and abnormality of the single heartbeat. For example, if 1 indicates that the single heartbeat is normal and 0 indicates that the single heartbeat is abnormal, then 25 single heartbeats are in 25 combinations of the series 0 and 1. It is possible. By combining the series, it is possible to recognize the heartbeat position where the abnormality occurs in the 25 single heartbeats.

As shown in FIG. 2C, the first convolution neural network includes four convolution layers, each of which convolves, includes processes such as convolution, activation and pooling. Each convolution layer has the same calculation order for each process, except that the input / output data and the convolution kernel size are different. That is, the single heartbeat data is input to the first convolutional neural network, then convolved and activated, and then pooled, thereby acquiring the feature data output from the convolutional layer. The convolutional layers of the first layer to the third layer include two-stage convolution, activation, and pooling in which each layer is continuous, while the convolutional layer of the fourth layer contains only one-stage convolution, activation, and pooling. include. In the activation calculation, functions such as sigmoid, tanH, and reLu are generally adopted. In the first convolutional neural network, the convolutional operation is used to extract the feature data of the electrocardiographic signal, the activation calculation is used to improve the non-linearity of the feature data, that is, the activity, and the pooling calculation is. , Used to reduce dimensions for feature data.

As shown in FIG. 2D, the second convolutional neural network includes three convolutional layers, and each convolutional layer has the same calculation order of each process except that the input / output data and the convolutional kernel size are different. That is, the input data (including the time-series data of the electrocardiographic signal) is input to the second convolutional neural network, then convolved and activated, and then the feature data output from the convolutional layer is acquired. The convolutional layers of the first layer to the third layer each include one-stage convolution and activation, but do not include pooling.

In this embodiment, the second convolutional neural network makes a pathological diagnosis for a continuous electrocardiographic signal, and at the same time, the first convolutional neural network can accurately identify the abnormal heartbeat position. Since the feature data output from the first convolutional neural network may be regarded as an approximation of the original single heartbeat, the feature data output from the first convolutional neural network can be integrated into the original electrocardiographic signal. The possibility of recognizing an electrocardiographic signal is greatly improved.

FIG. 3A is a schematic flowchart of an electrocardiographic signal detection method according to still another exemplary embodiment of the present invention, and FIG. 3B is an architecture for detecting an electrocardiographic signal applied to the embodiment shown in FIG. 3A. 3C is a schematic diagram of an architecture for detecting an electrocardiographic signal applied to an embodiment shown in FIG. 3A, which includes the following steps as shown in FIG. 3A.

In step 301, the electrocardiographic signal having a predetermined time length is divided to obtain a first predetermined number of single heartbeats.

The description of step 301 will not be described in detail here because the description of the embodiment shown in FIG. 1A may be referred to.

In step 302, by sequentially inputting the data of the first predetermined number of single heartbeats into the first convolutional neural network, the feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats is extracted.

The description of step 302 will not be described in detail here because the description of the embodiment shown in FIG. 2A may be referred to.

In step 303, the first predetermined number of feature data is input to the predetermined convolutional layer of the second convolutional neural network, and the electrocardiographic signal having a predetermined time length is input to the input layer of the second convolutional neural network, thereby performing the second convolution. The pathological type of the electrocardiographic signal is identified by a neural network.

The description of step 303 will not be described in detail here because the description of the embodiment shown in FIG. 2A may be referred to.

In the following, step 303 will be exemplified by combining FIGS. 3B and 3C. As shown in FIG. 3B, the data of the first predetermined number of single heartbeats obtained by division are sequentially input to the first convolutional neural network, and the original undivided electrocardiographic signal is input to the second convolutional neural network. You may enter it. For example, when the original continuous electrocardiographic signal is divided into 20 single heartbeats, the data of the 20 single heartbeats are sequentially input to the first convolutional neural network. After the 20 single heartbeat data enter the first convolutional neural network and perform the corresponding processing, the feature data output from the predetermined convolutional layer of the first convolutional neural network may be cached. .. For example, the feature data of 20 groups output from the convolutional layer of the second layer of the first convolutional neural network may be cached. When the amount of feature data output from the convolutional layer of the second layer for each single heartbeat is indicated as 21 * 32, the amount of feature data corresponding to 20 single heartbeats is 21 * 32 * 20 = 420 * 32. May be indicated.

The original continuous electrocardiographic signal is input to the input layer of the second convolutional neural network.

In one embodiment, the feature data obtained in the predetermined convolutional layer of the first convolutional neural network may be injected into the feature data obtained in the predetermined convolutional layer of the second convolutional neural network. As shown in FIG. 3C, the feature data obtained in the second convolutional layer of the first convolutional neural network is injected into the feature data obtained in the second convolutional layer of the second convolutional neural network, and this is injected. Both of the feature data of the two groups are input to the convolutional layer of the third layer of the second convolutional neural network. For example, when the amount of feature data obtained from the second convolutional layer of the second convolutional neural network is indicated as 531 * 32, the amount of feature data input by the third convolutional layer is 420 * 32 + 531 * 32 = ( 531-111) * 32 + 531 * 32 = 531 * 64-111 * 32. When the amount of feature data that can be processed by the convolution layer of the third layer is shown as 531 * 64, the convolution layer of the third layer lacks 111 * 32 feature data. With respect to the missing part, the zero padding method can ensure that the feature data actually input in the third layer convolution layer and the feature data required for processing the third layer convolution layer match. ..

What should be explained is that FIG. 3C is merely an exemplary explanation, and feature data of different dimensions can be output from a predetermined convolutional layer of the first convolutional neural network and injected into the predetermined convolutional layer of the second convolutional neural network. Therefore, the dependence of the second convolutional neural network on the feature data of the first convolutional neural network can be flexibly adjusted. The predetermined convolution layer here may be a certain convolution layer or a plurality of convolution layers. For example, the feature data output from the second convolutional layer of the first convolutional neural network is injected into the second convolutional layer of the second convolutional neural network, and the feature data output from the third convolutional layer of the first convolutional neural network. Can be injected into the third convolutional layer of the second convolutional neural network.

In this embodiment, the feature data output from the predetermined convolutional layer of the first convolutional neural network is input to the predetermined convolutional layer of the second convolutional neural network, and the recognition result of the second convolutional neural network is input to the first convolutional neural network. It is possible to make it dependent on the feature data of, and the recognition performance for the electrocardiographic signal can be improved.

FIG. 4A is a schematic flowchart of an electrocardiographic signal detection method according to another exemplary embodiment of the present invention, and FIG. 4B is an architecture for detecting an electrocardiographic signal applied to the embodiment shown in FIG. 4A. It is a schematic diagram of, and as shown in FIG. 4A, includes the following steps.

In step 401, an electrocardiographic signal having a predetermined time length is input to the second convolutional neural network, and the pathological type of the electrocardiographic signal is specified by the second convolutional neural network.

In step 402, when the pathology type indicates that an abnormality has occurred in the electrocardiographic signal, the electrocardiographic signal having a predetermined time length is divided to obtain a first predetermined number of single heartbeats.

In step 403, the acquired first predetermined number of single heartbeat data is input to the first convolutional neural network, and the first convolutional neural network identifies the heartbeat position where the abnormality occurs in the first predetermined number of single heartbeats. do.

Preferably, in step 403, the data of the first predetermined number of single heartbeats is input to the input layer of the first convolutional neural network, and each single heartbeat of the first predetermined number of single heartbeats is input by the first convolutional neural network. Is determined, the determination result is acquired, and the heartbeat position where the abnormality occurs in the first predetermined number of single heartbeats is specified based on the determination result.

As shown in FIG. 4B, data of a continuous electrocardiographic signal having a predetermined time length (for example, 30 seconds) is directly input to the second convolutional neural network, and the data of the continuous electrocardiographic signal is recognized to perform pathology. Get the type. If the pathology type indicates that an abnormality has occurred in the electrocardiographic signal, the electrocardiographic signal may be divided to obtain data of, for example, 25 single heartbeats. The data of these 25 single heartbeats are sequentially input to the first convolutional neural network and recognized, and the determination result is acquired. The determination result may be a series composed of 0 and 1. For example, if 0 indicates an abnormality and 1 indicates normal, 25 single heartbeats can correspond to a combination of 0s and 1s having a length of 25 bits. By recognizing the position where 0 is located, it is possible to identify the heartbeat position where the abnormality occurs in these 25 single heartbeats.

In this embodiment, when the pathological type recognized by the second convolutional neural network indicates that an abnormality has occurred in the electrocardiographic signal, the first convolutional neural network recognizes the single heartbeat of the electrocardiographic signal one by one. By doing so, the heartbeat position where the abnormality has occurred is recognized.

As can be seen from the architectural diagrams shown in FIGS. 2B, 3A, 3B and 4B above, the first convolutional neural network in the present invention is regarded as a single heart rate recognition network, and the second convolutional neural network is an electrocardiographic signal detection network. May be considered. The architectural design of the first convolutional neural network and the second convolutional neural network in the present invention has the following advantageous effects.
1) The first convolutional neural network can be trained independently to reduce the difficulty of training for the entire network architecture.
2) It is easy to acquire sufficient single heart rate data as a training sample of the first convolutional neural network. Therefore, the first convolutional neural network can be sufficiently trained, and the stability and reliability of recognition of a single heartbeat can be guaranteed.
3) Since the training of the second convolutional neural network can be reinforced based on the feature data obtained by the first convolutional neural network, the problem of data shortage of the electrocardiographic signal of a long series of continuous heartbeats can be solved.
4) Due to the small number of electrocardiographic signals, the architectural design in the present invention can be applied to embedded development applications.

FIG. 5 is a schematic structural diagram of an electrocardiographic signal detection device according to an exemplary embodiment of the present invention. As shown in FIG. 5, the electrocardiographic signal detection device may include a first division module 51, a first specific module 52, and a second specific module 53.

The first division module 51 divides an electrocardiographic signal having a predetermined time length to acquire a first predetermined number of single heartbeats.

The first specific module 52 specifies the feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats acquired by the first division module 51, and acquires the first predetermined number of feature data.

The second specific module 53 identifies the pathological type of the electrocardiographic signal having a predetermined time length based on the electrocardiographic signal having a predetermined time length and the first predetermined number of feature data specified by the first specific module 52.

FIG. 6 is a schematic structural diagram of an electrocardiographic signal detection device according to another exemplary embodiment of the present invention. As shown in FIG. 6, based on the embodiment shown in FIG. 5, the first specific module 52 may include a first input means 521 and an extraction means 522.

The first input means 521 sequentially inputs the data of the first predetermined number of single heartbeats to the first convolutional neural network.

The extraction means 522 extracts the feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats by the first convolutional neural network.

In one embodiment, the electrocardiographic signal detection device further includes a determination module 54 and a third specific module 55.

The determination module 54 makes a determination for each single heartbeat of the first predetermined number of single heartbeats acquired by the first division module 51 by the first convolutional neural network, and acquires the determination result.

The third specific module 55 identifies the heartbeat position where the abnormality occurs in the first predetermined number of single heartbeats based on the determination result acquired by the determination module 54.

In one embodiment, the second specific module 53 may include a first specific means 531, a second input means 532, and a second specific means 533.

The first specifying means 531 specifies the time-series data corresponding to each single heartbeat of the first predetermined number of single heartbeats, and acquires the first predetermined number of time-series data.

The second input means 532 inputs the first predetermined number of time series data acquired by the first specific means 531 and the first predetermined number of feature data to the input layer of the second convolutional neural network.

The second specifying means 533 identifies the pathological type of the electrocardiographic signal by the second convolutional neural network.

Specifically, the first specific means 531 is
For each single heartbeat in the electrocardiographic signal of a predetermined time length, the time point corresponding to the R wave of the single heartbeat is specified.
The time point corresponding to each R wave of the second predetermined number of single heartbeats adjacent to the R wave of the single heartbeat before and after is specified.
Corresponds to the single heartbeat based on the time point corresponding to the R wave of the single heartbeat and the time point corresponding to each R wave of the second predetermined number of single heartbeats adjacent to the single heartbeat before and after. Specify the time series data to be used.

In one embodiment, the second specific module 53 includes a third input means 534, a fourth input means 535, and a third specific means 536.

The third input means 534 inputs the first predetermined number of feature data to the predetermined convolutional layer of the second convolutional neural network.

The fourth input means 535 inputs an electrocardiographic signal having a predetermined time length to the input layer of the second convolutional neural network.

The third specific means 536 recognizes the first predetermined number of feature data input by the third input means 534 and the electrocardiographic signal input by the fourth input means 535 by the second convolutional neural network. , Identify the pathological type of electrocardiographic signal.

FIG. 7 is a schematic structural diagram of an electrocardiographic signal detection device according to still another exemplary embodiment of the present invention. As shown in FIG. 7, the electrocardiographic signal detection device may include a fourth specific module 71, a second division module 72, and a fifth specific module 73.

The fourth specific module 71 identifies the pathological type of the electrocardiographic signal having a predetermined time length by the second convolutional neural network.

The second division module 72 divides the electrocardiographic signal having a predetermined time length into a single predetermined number when the pathological type specified by the fourth specific module 71 indicates that an abnormality has occurred in the electrocardiographic signal. Get a heartbeat.

The fifth specific module 73 inputs the data of the first predetermined number of single heartbeats acquired by the second division module 72 into the first convolutional neural network, and the first convolutional neural network causes the first predetermined number of single heartbeats. Identify the heartbeat position where the abnormality occurred in one heartbeat.

In one embodiment, the fifth specific module 73 may include a fifth input means 731, a determination means 732, and a fourth specific means 733.

The fifth input means 731 inputs the data of the first predetermined number of single heartbeats to the input layer of the first convolutional neural network.

The determination means 732 makes a determination for each single heartbeat of the first predetermined number of single heartbeats input to the first convolutional neural network by the fifth input means 731, and acquires the determination result.

The fourth specifying means 733 identifies the heartbeat position where the abnormality occurs in the first predetermined number of single heartbeats based on the determination result acquired by the determination means 732.

The embodiment of the memory detection device of the present invention is applicable to electronic devices. The device embodiment may be implemented by software, or may be implemented by hardware or a combination of software and hardware. Taking software implementation as an example, a logically meaningful device is operated by the processor of an located electronic device reading the executable command of the corresponding device on a non-transient recording medium and transferring it to memory. It is formed. With respect to hardware implementation, FIG. 8 is a hardware structural diagram of the electronic device in which the electrocardiographic signal detector of the present invention is located, the processor 801 shown in FIG. 8, the memory 802, the network interface 803, and the non-transient In addition to the recording medium 804, the electronic device in which the device in the embodiment is located may usually be equipped with other hardware depending on the actual function of the electronic device. I will not repeat it here.

Those skilled in the art will readily come up with other embodiments of the invention after practicing the invention disclosed herein in light of the specification. The present invention is intended to cover any variation, use or adaptive variation of the invention. These variations, uses or adaptive changes, in accordance with the general principles of the invention, include common sense and conventional techniques in the art not disclosed in the invention. The specification and examples are considered merely exemplary and the true scope and gist of the invention is set forth by claim.

Further to be explained, it is intended that the terms "include", "provide" or any other synonym cover non-exclusive inclusion. Thus, a procedure, method, article or device having a series of elements not only has those elements, but also has other elements that are not explicitly listed, or such procedures, methods, articles. Or it also has elements specific to the device. Unless further restricted, the element limited by the phrase "contains one ..." does not deliberately exclude having the same other element in the procedure, method, article or device having the element.

The above are merely preferred embodiments of the invention and are not intended to limit the invention. Any changes, substitutions, improvements, etc. made within the spirit and principles of the invention are also included within the scope of the invention.

Claims (9)

An electrocardiographic signal detection method applied to electronic devices for detecting electrocardiographic signals.
A step of dividing an electrocardiographic signal of a predetermined time length to obtain a first predetermined number of single heartbeats, and
A step of identifying the feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats and acquiring the feature data of the first predetermined number of heartbeats.
Including a step of identifying the pathological type of the electrocardiographic signal of the predetermined time length based on the electrocardiographic signal of the predetermined time length and the feature data of the first predetermined number.
The step of identifying the feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats is
The data of the first predetermined number of single heartbeats are sequentially input to the first convolutional neural network, and
Including extracting feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats by the first convolutional neural network.
The step of identifying the pathological type of the electrocardiographic signal of the predetermined time length based on the electrocardiographic signal of the predetermined time length and the feature data of the first predetermined number is
Identifying the time-series data corresponding to each single heartbeat of the first predetermined number of single heartbeats, and acquiring the first predetermined number of time-series data.
By inputting the first predetermined number of time series data and the first predetermined number of feature data into the input layer of the second convolutional neural network, the pathological type of the electrocardiographic signal is specified by the second convolutional neural network. To do, including,
A method for detecting an electrocardiographic signal. A step of making a judgment for each single heartbeat of the first predetermined number of single heartbeats by the first convolutional neural network and acquiring the judgment result.
The electrocardiographic signal detection method according to claim 1, further comprising a step of identifying a heartbeat position in which an abnormality has occurred in the first predetermined number of single heartbeats based on the determination result. Identifying the time series data corresponding to the single heartbeat is
Identifying the time point corresponding to the R wave of the single heartbeat,
Identifying the time point corresponding to each R wave of the second predetermined number of single heartbeats adjacent to the R wave of the single heartbeat before and after.
Corresponds to the single heartbeat based on the time point corresponding to the R wave of the single heartbeat and the time point corresponding to each R wave of the second predetermined number of single heartbeats adjacent to the single heartbeat before and after. The electrocardiographic signal detection method according to claim 1, wherein the time-series data to be specified is specified. The step of identifying the pathological type of the electrocardiographic signal of the predetermined time length based on the electrocardiographic signal of the predetermined time length and the feature data of the first predetermined number is
Inputting the first predetermined number of feature data into the predetermined convolutional layer of the second convolutional neural network, and
By inputting the electrocardiographic signal having a predetermined time length to the input layer of the second convolutional neural network,
The electrocardiographic signal detection method according to claim 1, wherein the type of the electrocardiographic signal is specified by the second convolutional neural network. A first division module for dividing an electrocardiographic signal having a predetermined time length to obtain a first predetermined number of single heartbeats,
A first specific module for specifying the feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats acquired by the first division module and acquiring the first predetermined number of feature data, and
A second for identifying the pathological type of the electrocardiographic signal having a predetermined time length based on the electrocardiographic signal having a predetermined time length and the feature data of the first predetermined number specified by the first specific module. With a specific module,
The first specific module is
A first input means for sequentially inputting the data of the first predetermined number of single heartbeats into the first convolutional neural network, and
An extraction means for extracting feature data corresponding to each single heartbeat of the first predetermined number of single heartbeats by the first convolutional neural network is provided.
The second specific module is
A first specific means for specifying time-series data corresponding to each single heartbeat of the first predetermined number of single heartbeats and acquiring the first predetermined number of time-series data.
A second input means for inputting the first predetermined number of time series data acquired by the first specific means and the first predetermined number of feature data into the input layer of the second convolutional neural network.
An electrocardiographic signal detection device comprising: a second specifying means for identifying a pathological type of the electrocardiographic signal by the second convolutional neural network. A determination module for making a determination for each single heartbeat of the first predetermined number of single heartbeats by the first convolutional neural network and acquiring the determination result,
A claim characterized by further comprising a third specific module for identifying a heartbeat position in which an abnormality has occurred in the first predetermined number of single heartbeats based on the determination result acquired by the determination module. Item 5. The electrocardiographic signal detection device according to Item 5. The second specific module is
A third input means for inputting the first predetermined number of feature data into the predetermined convolutional layer of the second convolutional neural network, and
A fourth input means for inputting an electrocardiographic signal having a predetermined time length to the input layer of the second convolutional neural network, and
The second convolutional neural network recognizes the first predetermined number of feature data input by the third input means and the electrocardiographic signal input by the fourth input means, and recognizes the heart. The electrocardiographic signal detection device according to claim 5 , further comprising a third specific means for identifying a pathological type of an electric signal. A storage medium that can be read by a device
The device-readable storage medium is characterized in that an executable command of the device for executing the electrocardiographic signal detection method according to any one of claims 1 to 4 is stored. Device readable storage medium. It comprises a processor and a storage medium for storing the executable instructions of the processor.
The processor is an electronic device according to any one of claims 1 to 4 , wherein the processor executes the electrocardiographic signal detection method.

Images